A new approach to construct representative future forcing data for dynamic downscaling

Abstract

Climate downscaling using regional climate models (RCMs) has been widely used to generate local climate change information needed for climate change impact assessments and other applications. Six-hourly data from individual simulations by global climate models (GCMs) are often used as the lateral forcing for the RCMs. However, such forcing often contains both internal variations and externally-forced changes, which complicate the interpretation of the downscaled changes. Here, we describe a new approach to construct representative forcing for RCM-based climate downscaling and discuss some related issues. The new approach combines the transient weather signal from one GCM simulation with the monthly mean climate states from the multi-model ensemble mean for the present and future periods, together with a bias correction term. It ensures that the mean climate differences in the forcing data between the present and future periods represent externally-forced changes only and are representative of the multi-model ensemble mean, while changes in transient weather patterns are also considered based on one select GCM simulation. The adjustments through the monthly fields are comparable in magnitude to the bias correction term and are small compared with the variations in 6-hourly data. Any inconsistency among the independently adjusted forcing fields is likely to be small and have little impact. For quantifying the mean response to future external forcing, this approach avoids the need to perform RCM large ensemble simulations forced by different GCM outputs, which can be very expensive. It also allows changes in transient weather patterns to be included in the lateral forcing, in contrast to the Pseudo Global Warming (PGW) approach, in which only the mean climate change is considered. However, it does not address the uncertainty associated with internal variability or inter-model spreads. The simulated transient weather changes may also be unrepresentative of other models. This new approach has been applied to construct the forcing data for the second phase of the WRF-based downscaling over much of North America with 4 km grid spacing.